openai · OpenAI Platform Docs

Safety in building agents

Provides security best practices for building multi-agent workflows, including mitigating prompt injections, preventing private data leakage, and implementing structured outputs to constrain data flow.

Import to Prompt Buddy

Derived skill

Files assembled from official documentation

Viewing SKILL.md

Safety in building agents

When To Use

Use when designing or auditing multi-agent workflows to prevent prompt injection attacks, data exfiltration, and unintended tool execution.

Reference Files

File	Contains	Use For
`SKILL.md`	Entry point: scope, routing table, and workflow.	Start here.
`docs/safety-in-building-agents-workflow-guide.md`	A guide detailing risk types and mitigation strategies for building safe multi-agent workflows with OpenAI Agent Builder.	Questions about a guide detailing risk types and mitigation strategies for building safe multi-agent workflows with OpenAI Agent Buil...

What This Skill Covers

As you build and deploy agents with Agent Builder, it's important to understand the risks. Learn about risk types and how to mitigate them when building mult...
Main sections: Types of risk, Prompt injections, Private data leakage, Don't use untrusted variables in developer messages, Use structured outputs to constrain data flow.

Workflow

Open the most relevant file under docs/ for the exact documented workflow and wording.
Open schemas/ files for exact structured contracts.
Open examples/ files for concrete requests, commands, snippets, and manifests.
Do not add behavior or configuration that is not present in the attached source files.

Canonical source: https://developers.openai.com/api/docs/guides/agent-builder-safety.md